Thoughts on natural language processing and possible research

Hmm, needs more thought.

Wearable computing application: memory augmentation (personal information management, elderly)

problems facing wearable computing (re information extraction/retrieval):

- SPEECH-DRIVEN IMPLICIT QUERIES. A wearable computer can assist

human-human interaction by monitoring keywords in the conversation and suggesting relevant information, which will be presented in the background (visual, monophonic audio, spatially-located audio). A limited-vocabulary speaker-dependent speech engine recognizes names and keywords associated with personal information management while the wearer talks to other people. For example, the user might say, "Let me look at my schedule for Tuesday." The system will recognize "TUESDAY" and display a link for this Tuesday's schedule. The user can then key in further input or use structured voice commands to navigate through the data. To avoid confusion with the ongoing human-human conversation, human-computer interaction should be structured and clearly separate. (*Note*: I think I want to work on this for my PhD. Implicit queries for personal information management.)

- ONLINE INTERACTION, OFFLINE DICTATION. A wearable computer can be

used to record everything you say so that you can process it later. The speech stream from a directional microphone includes both dialogue with the wearable computer and dialogue with other entities. Online use involves limited-vocabulary speech recognition. The full speech stream can be stored on the hard disk and tagged on the fly with markers for later transcription using a large-vocabulary speech recognition engine. The post-processing of the text stream can be performed off-line and need not be real-time.

- EXPLICIT QUERIES: IR, TEXT SUMMARIZATION. Explicit information

retrieval with text summarization is necessary in order to deal with information overload. Both exact and fuzzy searches need to be supported. Topics may need to be mapped using an ontology in order to take advantage of communal knowledge or search through personal information. Output needs to be summarized; browsing through a webpage to find an answer is inefficient and requires too much attention. However, explicit queries against personal data stores cannot take advantage of the redundancy of the Web, and the wearable computer will need fuzzy search capabilities. (This is hard!)